Steganalysis Based on JPEG Compatibility
Jessica Fridrich*a, Miroslav Goljanb, Rui Dub
a
Center for Intelligent Systems, bDepartment of Electrical Engineering,
SUNY Binghamton, Binghamton, NY 13902-6000
ABSTRACT
In this paper, we introduce a new forensic tool that can reliably detect modifications in digital images, such as distortion
due to steganography and watermarking, in images that were originally stored in the JPEG format. The JPEG compression
leaves unique fingerprints and serves as a “fragile watermark” enabling us to detect changes as small as modifying the LSB
of one randomly chosen pixel. The detection of changes is based on investigating the compatibility of 8×8 blocks of pixels
with JPEG compression with a given quantization matrix. The proposed steganalytic method is applicable to virtually all
steganographic and watermarking algorithms with the exception of those that embed message bits into the quantized JPEG
DCT coefficients. The method can also be used to estimate the size of the secret message and identify the pixels that carry
message bits. As a consequence of our steganalysis, we strongly recommend avoiding using images that have been
originally stored in the JPEG format as cover-images for spatial-domain steganography.
Keywords: Steganography, steganalysis, JPEG
1. INTRODUCTION
Steganography is the art of secret communication. Its purpose is to hide the very presence of communication as opposed to
cryptography whose goal is to make communication unintelligible to those who do not posses the right keys1. Digital
images, videos, sound files, and other computer files that contain perceptually irrelevant or redundant information can be
used as “covers” or carriers to hide secret messages. After embedding a secret message into the cover-image, a so-called
stego-image is obtained. It is important that the stego-image does not contain any easily detectable artifacts due to message
embedding. A third party could use such artifacts as an indication that a secret message is present. Once this message
detection can be reliably achieved, the steganographic tool becomes useless.
Obviously, the less information is embedded into the cover-image, the smaller the probability of introducing detectable
artifacts by the embedding process. Another important factor is the choice of the cover-image. The selection is at the
discretion of the person who sends the message. The sender should avoid using cover-images that would be easy to analyze
for presence of secret messages. For example, one should not use computer art, charts, images with large areas of uniform
color, images with only a few colors, and images with a unique semantic content, such as fonts. Although computergenerated fractal images may seem as good covers6 because of their complexity and irregularity, they are generated by
strict deterministic rules that may be easily violated by message embedding3.
Scans of photographs or images obtained with a digital camera contain a high number of colors and are usually
recommended and considered safe for steganography. Some steganographic experts recommend grayscale images as the
best cover-images2.
There are essentially three types of image formats: raw, uncompressed formats (BMP, PCX), palette formats (GIF), and
lossy compressed formats (JPEG, Wavelet, JPEG2000). Only few current steganographic programs offer the capability to
embed messages directly in the JPEG stream. It is a difficult problem to devise a steganographic method that would hide
messages in the JPEG stream in a secure manner while keeping the capacity practical. Far more programs use the BMP,
PCX, or the GIF palette-based format. The GIF format is a difficult environment for secure steganography with reasonable
capacity3,7. Also, most steganographic techniques for GIFs implemented in current software products prioritize capacity
over security and are thus relatively easy to detect4,5. The raw formats, such as BMP, offer the highest capacity and best
*
Correspondence: Email: [email protected]; WWW: http://www.ssie.binghamton.edu/fridrich; Telephone: 607
777 2577; Fax: 607 777 2577
overall security. In this paper, we demonstrate that even 24-bit images or grayscale 8-bit images may actually be extremely
poor candidates for cover-images if they were initially acquired as JPEG images and later decompressed to a lossless
format. In fact, it is quite reasonable to expect that most casual users of steganographic programs will use scanned images
or images from a digital camera that were originally stored in the JPEG format due to its efficiency in data storage.
All steganographic methods strive to achieve the minimal amount of distortion in order to minimize the likelihood of
introducing any visible artifacts. Consequently, if the cover-image, was initially stored in the JPEG format, the act of
message embedding will not erase the characteristic structure created by the JPEG compression and one can still easily
determine whether or not a given image has been stored as JPEG in the past. Actually, unless the image is too small, one
can reliably recover even the values of the JPEG quantization table by carefully analyzing the values of DCT coefficients
in all 8×8 blocks. After message embedding, however, the cover-image will become (with a high probability) incompatible
with the JPEG format in the sense that it may be possible to prove that a particular 8×8 block of pixels could not have been
produced by JPEG decompression of any block of quantized coefficients. This finding provides strong evidence that the
block has been modified. It is highly suspicious to find an image stored in a lossless format that bears a strong fingerprint
of JPEG compression, yet is not fully compatible with any JPEG compressed image. This can be interpreted as evidence
for steganography.
By checking the JPEG compatibility of every block, we can potentially detect messages as short as one bit. And the
steganalytic method will work for virtually any steganographic or watermarking method, not just the LSB embedding!
Indeed, in our experiments, we have found out that even one randomly selected pixel whose gray level has been modified
by one can be detected with very high probability. For longer messages, one can even attempt to estimate the message
length and its position in the image by determining which 8×8 blocks are incompatible with JPEG compression. It is even
possible to analyze the image and estimate the likely candidate for the cover-image or its blocks (the "closest" JPEG
compatible image/block). This way, we may be able to identify individual pixels that have been modified. All this indicates
that an extremely serious information leakage from the steganographic method can occur and thus completely compromise
the steganographic channel.
In this paper, we elaborate on the idea presented in the previous paragraph. In Section 2, we describe an algorithm that can
decide if a given 8×8 block of pixels is compatible with JPEG compression with a given quantization matrix. Appendix A
contains details on how the quantization matrix can be estimated from the image. Throughout the paper, we point out some
limitations of the proposed steganalytic technique and finally, in Section 3, we conclude the paper and outline future
research directions. As a consequence of this research, we strongly urge users of steganographic programs to avoid images
previously stored in the JPEG format as cover-images.
2. STEGANALYSIS BASED ON JPEG COMPATIBILITY
Although in this paper we explain the technique on grayscale images, it can be extended to color images in a
straightforward manner. We start with a short description of the JPEG compression algorithm. In JPEG compression, the
image is first divided into disjoint blocks of 8×8 pixels. For each block Borig (with integer pixel values in the range 0−255),
the discrete cosine transform (DCT) is calculated, producing 64 DCT coefficients. Let us denote the i-th DCT coefficient of
the k-th block as dk(i), 0 ≤ i ≤ 64, k = 1, …, T, where T is the total number of blocks in the image. In each block, all 64
coefficients are further quantized to integers Dk(i) using the JPEG quantization matrix Q
 d (i) 
Dk (i) = integer _ round  k  .
 Q(i) 
The quantized coefficients Dk(i) are arranged in a zig-zag manner and compressed using the Huffman coder. The resulting
compressed stream together with a header forms the final JPEG file.
The decompression works in the opposite order. The JPEG bit-stream is decompressed using the Huffman coder and the
quantized DCT coefficients Dk(i) are multiplied by Q(i) to obtain DCT coefficients QDk, QDk(i) = Q(i)Dk(i) for all k and i.
Then, the inverse DCT is applied to QDk and the result is rounded to integers in the range 0−255
B = [Braw ] , where Braw = DCT −1 (QD) ,
(1)
and [x]=integer_round(x) for 0 ≤ x ≤ 255, [x]=0 for x < 0, and [x]=255 for x > 255. In the last expression, we dropped the
block index k to simplify the notation. We note that because the JPEG compression is lossy, in general Borig may not be
equal to B.
If the block B has no pixels saturated at 0 or 255, we can write in the L2 norm
||Braw−B||2 ≤ 16,
(2)
because |Braw(i)−B(i)|≤1/2 for all i = 1, …, 64 due to rounding.
Suppose that we know the quantization matrix Q (see Appendix A). Our steganalytic technique is based on the following
question:
Given an arbitrary 8×8 block of pixel values B, could this block have arisen through the process of JPEG
decompression with the quantization matrix Q?
Denoting QD'=DCT(B), we can write using the Parserval's equality
B − Braw
2
= DCT ( B ) − DCT ( Braw )
2
= QD' −QD
2
≤ 16 .
On the other hand, we can find a lower estimate for the expression ||QD’−QD||2 by substituting for QD(i) the closest integer
multiple of Q(i):
16 ≥ QD'−QD
2
≥
 QD' (i ) 
 =S.
Q(i ) 
64
∑ QD' (i) − Q(i)round 
i =1
(3)
The quantity S can be calculated from the block B provided the quantization matrix Q is known. If S is larger than 16, we
can conclude that the image block B is not compatible with JPEG compression with the quantization matrix Q. We reiterate
that this is true only for blocks that do not have pixels that are saturated at 0 or 255. Indeed, the estimate (2) may not hold
for blocks that have saturated pixels because the rounding at 0 and 255 can be much larger than 1/2.
If, for a given block B with unsaturated pixels, S is smaller than 16, the block B may or may not be JPEG compatible. Let
qp(i), p=1, … , be integer multiples of Q(i) that are closest to QD(i) ordered by their distance from QD(i) (the closest
multiple is q1(i)). In order to decide whether or not a given block B is compatible with JPEG compression with quantization
table Q, we need to inspect all 64-tuples of indices {p(1), …, p(64)} for which
S=
64
∑ QD' (i) − q
p (i )
(i ) ≤ 16
(4)
i =1
and check if
B=[DCT−1(QD)], where QD(i)= qp(i)(i).
(5)
If, for at least one set of indices {p(1), …, p(64)}, the expression (5) is satisfied, the block B is JPEG compatible, otherwise
it is not.
The number of 64-tuples {p(1), …, p(64)} satisfying expression (4) is always finite but it rapidly increases with increasing
JPEG quality factor. For quality factors higher than 95, a large number of quantization factors Q(i) become 1 or 2, and the
total number of combinations of all 64 indices becomes too large to handle. We performed our experiments in Matlab on a
Pentium II computer with 128MB memory. Once the quality factor exceeded 95, the running time became too long because
Matlab ran out of memory and had to access the hard disk. We acknowledge this complexity increase as a limitation of our
approach. In the future, we would like to develop a better and faster algorithm for testing JPEG compatibility for high
quality compression factors.
Description of the algorithm:
1.
2.
3.
4.
5.
6.
7.
Divide the image into a grid of 8×8 blocks, skipping the last few rows or columns if the image dimensions are not
multiples of 8.
Arrange the blocks in a list and remove all saturated blocks from the list (a block is saturated if it has at least one pixel
with a gray value 0 or 255). Denote the total number of blocks as T.
Extract the quantization matrix Q from all T blocks as described in Appendix A. If all the elements of Q are ones, the
image was not previously stored as JPEG and our steganalytic method does not apply (exit this algorithm). If more
than one plausible candidate exists for Q, the steps 4−6 need to be carried out for all candidates and the results that
give the highest number of JPEG compatible blocks will be accepted as the result of this algorithm.
For each block B calculate the quantity S (see equation (3)).
If S>16, the block B is not compatible with JPEG compression with quantization matrix Q. If S≤16, for each DCT
coefficient QDi' calculate the closest multiples of Q(i), order them by their distance from QDi', and denote them qp(i),
p=1, …. For those combinations, for which the inequality (4) is satisfied, check if expression (5) holds. If, for at least
one set of indices {p(1), …, p(64)} the expression (5) is satisfied, the block B is JPEG compatible, otherwise it is not.
After going through all T blocks, if no incompatible JPEG blocks are found, the conclusion is that our steganalytic
method did not find any evidence for presence of secret messages. If, on the other hand, there are some JPEG
incompatible blocks, we can attempt to estimate the size of the secret message, locate the message-bearing pixels, and
even attempt to obtain the original cover image before secret message embedding started.
If all blocks are identified as JPEG incompatible or if the image does not appear to be previously stored as JPEG, we
should repeat the algorithm for different 8×8 divisions of the image (shifted by 0 to 7 pixels in the x and y directions).
This step may be necessary if the cover image has been cropped prior to message embedding.
3. CONCLUSIONS AND FUTURE EFFORT
In this paper, we describe a new steganographic technique that can reliably detect modifications in digital images, such as
those due to steganography and watermarking, as long as the original image (the cover-image) has been previously stored
in the JPEG format. The steganalytic technique starts with extracting the JPEG quantization matrix by carefully inspecting
the clusters of DCT coefficients in all 8×8 blocks. Then, each block is analyzed if its pixel values are truncated values of
an inverse DCT transform of a set of coefficients quantized with the extracted quantization matrix. A simple necessary
condition is derived that makes the analysis computationally feasible. If the corresponding quantized DCT coefficients are
found, the block is termed compatible, otherwise it is not.
The steganalytic technique will work for all steganographic methods, except the methods that embed information directly
into the compressed JPEG stream. It has a potential to detect changes as small as one pixel. By inspecting the closest
compatible JPEG block, we can even attempt to locate the pixels that have been modified.
As a consequence of this research, we strongly urge the users of steganographic programs to avoid using images that have
been previously stored in the JPEG format as cover-images for steganography. The JPEG compression imprints a unique
fingerprint on the image and this fingerprint can be carefully utilized for steganalysis as shown in this paper. If no other
images are available, try to rescale or resample the image to slightly smaller dimensions to wipe out the JPEG fingerprint.
Applying filters, adjusting contrast, or slightly rotating the image may also help in reducing the JPEG fingerprint.
The proposed steganalytic method has some limitations that we would like to address in the future. First, the necessary
condition derived for unsaturated blocks (3) does not hold for blocks that have some black or white pixels. The study of
JPEG compatibility of blocks that have some saturated pixels appears to be a difficult challenge that will be pursued in the
future.
Another limitation of our technique is the rapidly increasing computational complexity with increasing JPEG quality factor.
The running time became unacceptably large for quality factors larger than 95. This is due to the large number of candidate
JPEG blocks that need to be inspected for compatibility. We believe that this complexity problem could be resolved by
deriving a set of other necessary conditions similar to (3) to decrease the number of possible candidates that need to be
inspected for compatibility in Step 5.
APPENDIX A
In this appendix, we show how one can extract the quantization matrix Q from a (BMP) image. Keeping the same notation
as in the previous section, we first calculate the DCT coefficients dk(i), 0 ≤ i ≤ 64, k = 1, …, T from all unsaturated 8×8
blocks. For each coefficient i, we plot the quantity Ei(q) as a function of q = 1, 2, …
Ei (q) =
1
T
T
∑d
k =1
k
 d (i ) 
(i ) − q × integer _ round  k  .
 q 
Figure 1 The error E48(q) for the 48-th DCT coefficient (8,6) plotted as a function of the quantization step q. Notice the local minima at
30 and all its divisors 15, 10, 6, 5, 3, and 2. The last minimum is 30, which corresponds to the correct quantization step for JPEG quality
factor 85.
The quantity Ei(q) measures the compatibility of all i-th DCT coefficients in the image with the quantization step q. If the
image under inspection was indeed previously stored as JPEG, we will observe a drop (local minimum) at the correct
quantization value q and at all integer divisors of q. It is intuitively clear that the correct value of the quantization step
should be the largest value q at which a local minimum of E occurs. As discussed below, this may not, however, be true in
all cases.
First of all, it is possible that some "ghost" local minima can occur for values of q larger than the correct value. Those
minima, however, are significantly larger than the other “correct” minima. By setting a threshold on the largest possible
value of a local minimum, we can successfully “filter out” the false minima. In our experiments, we make the thresholod
dependent on the DCT coefficient (on index i) and calculate its value as µi+3σi, where µi and σi are the mean and the
standard deviation of all local minima of the vector Ei(q), q=1, ….
If the vector Ei(q) does not have any local minima except for q=1 (we note that Ei(1)≤ Ei(q) for all q>1), we inspect the
difference between Ei(1) and Ei(2) to decide if the quantization step is 1 or 2. For this case, we developed an empirical rule
that gives very good results. Based on our experiments, if Ei(1) < 0.6 Ei(2), we conclude that the quantization step is 1,
otherwise it is 2.
This process of finding the quantization steps for each coefficient is still not completely foolproof. For small images, it can
happen that, for the quantization step q, all values of the DCT coefficients dk(i) will be multiples of 2q. In this case, we
have no means to tell if the correct step is 2q or q. The probability of this happening decreases with the image size. To
address this problem, we inspect the extracted quantization matrix Q for outliers by dividing it by the standard JPEG
quantization table. Here, we use the logic that even customized tables (e.g., tables used in JPEG compression in digital
cameras) will not significantly deviate from the standard quantization table. Finally, it is always possible to identify
candidates for suspicious quantization steps and run the steganalysis for all possible combinations.
ACKNOWLEDGEMENTS
The work on this paper was supported by Air Force Research Laboratory, Air Force Material Command, USAF, under a
research grant number F30602-00-1-0521. The U.S. Government is authorized to reproduce and distribute reprints for
Governmental purposes notwithstanding any copyright notation there on. The views and conclusions contained herein are
those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or
implied, of Air Force Research Laboratory, or the U. S. Government.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
Andersen, R.J. and Petitcolas, F.A.P., "On the limits of steganography", IEEE Journal of Selected Areas in
Communications, Special Issue on Copyright and Privacy Protection, 16(4), pp. 474−481, 1998.
Aura, T., “Invisible communication”, Proc. of the HUT Seminar on Network Security ‘95, Espoo, Finland, November
1995. Telecommunications Software and Multimedia Laboratory, Helsinki University of Technology.
Fridrich, J. and Du, R., “Secure Steganographic Methods for Palette Images”, Proc. The 3rd Information Hiding
Workshop, September 28−30, Dresden, Germany, LNCS vol. 1768, Springer-Verlag, New York, pp. 47−60, 1999.
Johnson, N. F. and Jajodia, S., “Steganography: Seeing the Unseen,” IEEE Computer, February, pp.26−34, 1998.
Johnson, N. F. and Jajodia, S., “Steganalysis of Images Created Using Current Steganography Software,” Proc. The
2nd Information Hiding Workshop, Portland, OR, April, LNCS vol.1525, Springer-Verlag, New York, 1998.
Hastur, H., “Mandelsteg,” Software downloadable from http://idea.sec.dsi.uimi.it/pub/security/crypt/codev, 1994.
Westfeld, A. and Pfitzmann, A., "Attacks on Steganographic Systems", Proc. The 3rd Information Hiding Workshop,
September 28−30, Dresden, Germany, LNCS vol. 1768, Springer-Verlag, New York, pp. 61−75, 1999.